AI Sdk UI

AI SDK UI - Chat & Generative UI Framework

AI SDK UI provides framework-agnostic hooks for building interactive chat, completion, and assistant applications with real-time streaming and state management.

Quick Start: Basic Chat

Client (React)

'use client';
import { useChat } from '@ai-sdk/react';
import { DefaultChatTransport } from 'ai';
import { useState } from 'react';

export default function Chat() {
  const { messages, sendMessage, status } = useChat({
    transport: new DefaultChatTransport({ api: '/api/chat' }),
  });
  const [input, setInput] = useState('');

  return (
    <>
      {messages.map(message => (
        <div key={message.id}>
          {message.role === 'user' ? 'User: ' : 'AI: '}
          {message.parts.map((part, index) =>
            part.type === 'text' ? <span key={index}>{part.text}</span> : null
          )}
        </div>
      ))}

      <form => {
        e.preventDefault();
        if (input.trim()) {
          sendMessage({ text: input });
          setInput('');
        }
      }}>
        <input
          value={input}
          => setInput(e.target.value)}
          disabled={status !== 'ready'}
        />
        <button type="submit" disabled={status !== 'ready'}>
          Submit
        </button>
      </form>
    </>
  );
}

Server (Next.js App Router)

import { convertToModelMessages, streamText, UIMessage } from 'ai';
import { openai } from '@ai-sdk/openai';

export const maxDuration = 30;

export async function POST(req: Request) {
  const { messages }: { messages: UIMessage[] } = await req.json();

  const result = streamText({
    model: openai('gpt-4o'),
    system: 'You are a helpful assistant.',
    messages: await convertToModelMessages(messages), // v6: now async
  });

  return result.toUIMessageStreamResponse();
}

Hook Selection Guide

Decision tree:

• Need conversation history + tools? → useChat
• Need typed/structured streaming data? → useObject
• Need simple text completion? → useCompletion

Core Patterns

1. Status Management

const { status, stop } = useChat();

// status values: 'submitted' | 'streaming' | 'ready' | 'error'

{(status === 'submitted' || status === 'streaming') && (
  <div>
    {status === 'submitted' && <Spinner />}
    <button => stop()}>Stop</button>
  </div>
)}

2. Error Handling

const { error, reload } = useChat();

{error && (
  <>
    <div>An error occurred.</div>
    <button => reload()}>Retry</button>
  </>
)}

Server-side error messages:

return result.toUIMessageStreamResponse({
  onError: error => {
    if (error instanceof Error) return error.message;
    return 'Unknown error';
  },
});

3. Message Modification

const { messages, setMessages } = useChat();

const handleDelete = (id: string) => {
  setMessages(messages.filter(m => m.id !== id));
};

const handleEdit = (id: string, newText: string) => {
  setMessages(messages.map(m =>
    m.id === id
      ? { ...m, parts: [{ type: 'text', text: newText }] }
      : m
  ));
};

4. File Attachments

const [files, setFiles] = useState<FileList | undefined>();

<form => {
  e.preventDefault();
  sendMessage({ text: input, files });
  setFiles(undefined);
}}>
  <input
    type="file"
    => setFiles(e.target.files ?? undefined)}
    multiple
  />
</form>

5. Custom Request Options

// Per-request customization (recommended)
sendMessage(
  { text: input },
  {
    headers: { Authorization: 'Bearer token' },
    body: { temperature: 0.7, user_id: '123' },
    metadata: { sessionId: 'abc' },
  }
);

// Hook-level configuration
const { sendMessage } = useChat({
  transport: new DefaultChatTransport({
    api: '/api/chat',
    headers: () => ({ Authorization: `Bearer ${getToken()}` }),
    body: { systemContext: 'expert' },
  }),
});

6. Message Metadata

// Server: Attach metadata
return result.toUIMessageStreamResponse({
  messageMetadata: ({ part }) => {
    if (part.type === 'start') {
      return { createdAt: Date.now(), model: 'gpt-4o' };
    }
    if (part.type === 'finish') {
      return { totalTokens: part.totalUsage.totalTokens };
    }
  },
});

// Client: Access metadata
{messages.map(m => (
  <div key={m.id}>
    {m.metadata?.createdAt && new Date(m.metadata.createdAt).toLocaleString()}
    {m.parts.map(part => part.type === 'text' ? part.text : null)}
    {m.metadata?.totalTokens && <span>{m.metadata.totalTokens} tokens</span>}
  </div>
))}

7. Regenerate & Stop

const { regenerate, stop, status } = useChat();

<>
  <button disabled={status !== 'streaming'}>
    Stop
  </button>
  <button disabled={!(status === 'ready' || status === 'error')}>
    Regenerate
  </button>
</>

Message Parts Type Reference

Messages use a parts array instead of content for flexible multi-modal rendering:

type MessagePart =
  | { type: 'text'; text: string }
  | { type: 'file'; filename: string; mediaType: string; url: string }
  | { type: 'tool-invocation'; toolName: string; input: unknown; result?: unknown }
  | { type: 'tool-result'; toolName: string; result: unknown }
  | { type: 'reasoning'; text: string }  // DeepSeek R1, Claude 3.7 Sonnet
  | { type: 'source-url'; id: string; url: string; title?: string }  // Perplexity, Google
  | { type: 'source-document'; id: string; title?: string };

// Render pattern
{message.parts.map((part, index) => {
  switch (part.type) {
    case 'text':
      return <span key={index}>{part.text}</span>;
    case 'file':
      return part.mediaType.startsWith('image/')
        ? <img key={index} src={part.url} alt={part.filename} />
        : null;
    case 'reasoning':
      return <pre key={index}>{part.text}</pre>;
    case 'source-url':
      return <a key={index} href={part.url}>{part.title ?? 'Source'}</a>;
    case 'tool-invocation':
      return <ToolUI key={index} tool={part} />;
    default:
      return null;
  }
})}

Offline Stack Scanner

Run the local scanner before chat UI migrations or reviews:

python3 skills/ai-sdk-ui/scripts/ai_stack_scan.py --root <repo> --pretty

It emits ai_stack_scan.v1, uses no network by default, skips symlinks, and flags likely UI migration risks such as message.content, legacy useChat helpers, missing @ai-sdk/react dependencies, and obsolete stream response helpers. Verify each signal against current AI SDK docs/source before editing. Keep full scanner JSON local; share only specific redacted signals externally.

Framework Support

AI Elements (shadcn/ui Components)

Pre-built UI components for chat interfaces: https://ai-sdk.dev/elements

Includes: Message bubbles, input fields, tool UIs, and more.

Reference Navigation

Event Callbacks

const { messages } = useChat({
  onFinish: ({ message, messages, isAbort, isDisconnect, isError }) => {
    // Log completion, update analytics, trigger side effects
    if (!isError) logMessage(message);
  },
  onError: error => {
    // Custom error handling, fallback UI
    Sentry.captureException(error);
  },
  onData: data => {
    // Process data parts, validate responses
    // Throw error to abort processing
  },
});

Advanced: Custom Transport

const { sendMessage } = useChat({
  transport: new DefaultChatTransport({
    prepareSendMessagesRequest: ({ id, messages, trigger, messageId }) => {
      if (trigger === 'submit-user-message') {
        return {
          body: {
            id,
            message: messages[messages.length - 1],
            messageId,
          },
        };
      }
      // Handle regenerate, custom triggers
    },
  }),
});

Type Inference for Tools

import { InferUITools, InferAgentUIMessage, ToolSet, UIMessage } from 'ai';

const tools = {
  weather: {
    description: 'Get weather',
    inputSchema: z.object({ location: z.string() }),
    execute: async ({ location }) => `Sunny in ${location}`,
  },
} satisfies ToolSet;

type MyUITools = InferUITools<typeof tools>;
type MyUIMessage = UIMessage<never, never, MyUITools>;
// Or for agent messages:
// type MyAgentUIMessage = InferAgentUIMessage<typeof myAgent>;

const { messages } = useChat<MyUIMessage>();

Reasoning & Sources

// Enable reasoning (DeepSeek R1, Claude 3.7 Sonnet)
return result.toUIMessageStreamResponse({ sendReasoning: true });

// Enable sources (Perplexity, Google)
return result.toUIMessageStreamResponse({ sendSources: true });

// Render reasoning and sources
{message.parts.map(part => {
  if (part.type === 'reasoning') return <pre>{part.text}</pre>;
  if (part.type === 'source-url') return <a href={part.url}>{part.title}</a>;
})}

Performance: Throttle Updates

const { messages } = useChat({
  experimental_throttle: 50, // React only: throttle to 50ms
});

Plain Text Streams

import { TextStreamChatTransport } from 'ai';

const { messages } = useChat({
  transport: new TextStreamChatTransport({ api: '/api/chat' }),
});

Note: Tools, usage, and finish reasons unavailable with TextStreamChatTransport.